Singular radial solutions for the Keller-Segel equation in high dimension (Q2286496)

The authors analyze the behavior around the origin of singular and radial solutions to the steady Keller-Segel equation \(-\Delta v+v=\lambda e^{v}\) posed in \(B_{R}\setminus \{0\}\), with the constraint \(v>0\) and satisfying the boundary condition \(\partial _{\nu }v=0\) on \(\partial B_{R}\). Here \(B_{R}\) is the ball of \(\mathbb{R}^{N}\), \(N\geq 3\), centered at the origin and of radius \(R\). The authors also condider the equation \(-\Delta u+u=e^{\mu (u-1)}\) posed in \(B_{R}\setminus \{0\}\), with the constraint \(u>0\) and satisfying the boundary condition \(\partial _{\nu }u=0\) on \(\partial B_{R}\), whose branches were analyzed by \textit{D. Bonheure} et al. [Calc. Var. Partial Differ. Equ. 56, No. 3, Paper No. 74, 35 p. (2017; Zbl 1375.35097)]. Each branch was proved to decompose in two connected components. In the present paper, the authors will focus on the upper part of the branch of the above Keller-Segel equation. They recall the results obtained in the case of Dirichlet boundary conditions. The first result of the paper proves that for \(N\geq 3\) and \(\lambda >0\) there exists \(U_{\lambda }^{\ast }=U^{\ast }>0\) satisfying, for each \(\delta \in (0,1)\), the equation \(-u^{\prime \prime }-\frac{N-1}{N}u^{\prime }+u=\lambda e^{u}\) on \(\mathbb{R}^{+}\) with the condition \(u(r)=-2\ln r+\ln \frac{2(N-2)}{\lambda }+O(r^{2\delta })\) when \(r\rightarrow 0\). A solution satisfying the preceding equation with the condition \(u(r)=-2\ln r+\ln \frac{2(N-2)}{\lambda }+O(1)\) is unique. The solution of \(-u^{\prime \prime }-\frac{N-1}{N}u^{\prime }+u=\lambda _{n}e^{u}\) on \(\mathbb{R}^{+}\) with the boundary conditions \(u(0)=\gamma _{n}\), \(u^{\prime }(0)=0\) converges to \(U^{\ast }\) solution of the preceding problem for any sequences \((\gamma _{n})_{n}\) and \((\lambda _{n})_{n}\) respectively converging to \(+\infty \) and \(\lambda_{\infty }\in (0,\infty )\), if \(\lambda _{\infty }\) is less than some value \(\lambda _{N}^{\ast }\) which depends on \(N\). The first main result of the long paper proves that for \(N\geq 3\) and \(\lambda \in (0,\lambda _{N}^{\ast})\), there exists \(i^{\ast }\) such that for any \(i\geq i^{\ast }\) there exists \(\lambda ^{i}>0\) such that the above Keller-Segel equation admits a singular radial solution \(U_{\lambda ^{i}}^{\ast }\) satisfying \(\sharp \{r\in (0,1)\mid U_{\lambda ^{i}}^{\ast }(r)=\overline{u}_{\lambda ^{i}}\}=i\), where \(\overline{u}_{\lambda ^{i}}\) is the largest solution of \(u=\lambda ^{i}e^{u}\). The second main result proves that for \(3\leq N\leq 9\), \(R>0\), \(i\geq i^{\ast }\), there exists a sequence of initial data \((\gamma _{n})_{n}\) of the preceding problem with \(\gamma _{n}\rightarrow \infty \) and a sequence of positive integers \((j_{n})_{n}\) such that \(r_{\lambda ^{i},\gamma _{n}}^{j_{n}}=R\), with \(u^{\prime }(r_{\lambda ^{i}, \gamma_{n}}^{j_{n}},\gamma _{n})=0\). This proves that for each \(i\geq 1\), the branch \(\mathcal{B}_{i}^{+}\) oscillates around \(\lambda ^{i}\) when \(3\leq N\leq 9\). For their next main theorem, the authors introduce the Morse index \(m(v)\) of a solution \(v\) of the Keller-Segel equation in the space of radial functions as the number of negative eigenvalues \(\alpha \) (counting multiplicities) of the eigenvalue problem \(-\Delta \varphi +\varphi -\lambda e^{v}\varphi =\alpha \varphi \) in \(B_{R}\), \(\partial _{\nu }\varphi =0\) on \( \partial B_{R}\), \(\varphi \) is radially symmetric. They prove that If \(U_{\lambda ^{i}}^{\ast }\) is a solution to the Keller-Segel equation, then \(m(U_{\lambda ^{i}}^{\ast })<\infty \) when \(N>10\) while \(m(U_{\lambda^{i}}^{\ast })=\infty \) when \(3\leq N\leq 9\). The authors also prove that if there exists a singular solution on \(B_{R}\) for certain \(\lambda ^{\ast }\), then there is no singular solution for nearby \(\lambda \), that is the set \((\lambda ^{i})\) is discrete. They indeed prove the existence of a set \(S^{\ast }\subset (0,\infty )\) of Lebesgue measure zero, such that for any radius \(R\in (0,\infty )\setminus S^{\ast }\) the following holds. If \((U_{\lambda ^{\ast }}^{\ast })^{\prime }(R)=0\), then there exists \(\delta >0\) such that for any \(\lambda \in (\lambda ^{\ast }-\delta ,\lambda ^{\ast}+\delta )\setminus \{\lambda ^{\ast }\}\) one has \((U_{\lambda }^{\ast})^{\prime }(R)\neq 0\). The last main result proves that for \(N\geq 4\) and any \(\lambda <1/e\) for \(N>4\) and \(\lambda <0.196\) if \(N=4\), and \(R\in (0,\infty )\setminus S^{\ast }\), there exist \(\delta >0\) and \(\Gamma >0\) such that for each \(\gamma \geq \Gamma \) there exists at most one \(\lambda \in (\lambda ^{i}-\delta ,\lambda ^{i}+\delta )\) such that \(r_{\lambda ,\gamma }^{i}=R\). For the proofs, the authors mainly use a Lyapunov functional associated to the Keller-Segel equation, a Pohozaev-type identity, the Sturm-Picone comparison theorem, Sard's theorem and previous results by \textit{D. D. Joseph} and \textit{T. S. Lundgren} in [Arch. Ration. Mech. Anal. 49, 241--269 (1973; Zbl 0266.34021)].